[00:00] You got any questions or anything from you or you're just like, go on, do it? Yeah, just do it.
[00:12] You are laughing and smiling, which comes to my first question I wanted to ask you, which is, how are you feeling at the moment? Better than a few weeks ago. I mean, I'm still having a little calm down.
[00:26] But certainly, you know, I mean, the first week. I had a two-month-old baby.
[00:38] The week that everything happened, I probably averaged two hours sleep every night. Some people would constantly do nothing on my office door.
[00:52] Sometimes the door was open and they'd walk in the house just to sleep at the desk. I was just shut the door. Does it feel like you've been mugged by 10,000 agents or something? Does it feel like that?
[01:05] Like you've been a victim or something from your perspective? Yeah, I mean, it's just so silly. I mean, it's like I'm not... I'm someone that is sort of aware of their culture, you know,
[01:20] because I have been dealing with sort of... I had this collaboration previously with Google DeepMind, so I kind of knew how things worked within the tech industry,
[01:36] and I knew where anything goes, basically. So I knew the philosophy of move fast, break things.
[01:49] And that philosophy with math clashes. I use a thing that's been broken this time I mean the funny thing is like how did this thing
[02:02] it could have been so simple right if they just played their cards the way they should have played it and they'd let us release our results and then
[02:14] they launched their swarm then they had then the path would have been clear and the story would have been clear I mean, we would have got the Mavius program. LaVange worked for Anthropic, but this was his sort of side project with Matt,
[02:29] and he was working with me. This wasn't some, you know, multi-million dollar project of Anthropic at all. And so, sure, had, like, Dario said, let's beat OpenAI,
[02:45] then there'll be a different story. But it wasn't like that, and we didn't want it ever to be like that. So had they just let us release the result and then took that and then worked from there,
[03:00] and then they would have got the... It wouldn't have been such a big drama. Can I just get an idea from you, how long you have been on this collision course with Navier-Stokes?
[03:14] How long you've been on this particular journey? And was Navier-Stokes, this famous millennium problem, like a finish line or a goal or a milestone you were hoping to pass? Like, how long has this been building and where was it in your head before all the controversy?
[03:31] It's a great question. So it was never like I was never striving to solve a math problem.
[03:43] But it was like a north star for my entire career. So I saw it as like it was a problem out there.
[03:55] And this is how science and math work, is that you have these big problems out there that you'd love to solve, but they're unreachable.
[04:10] So what you do is you just put it out there, and you think of, what other similar problems can I do that will help me along that path towards that goal? So, that's basically, I haven't, all these years I haven't been working directly on solving the Madness
[04:26] Stokes. I have been spending the last, I don't know, six, seven years, directly on Singularity.
[04:38] But the Madness Stokes problem has always been for this North Star. And it was never my goal to solve it. It was always my goal to just be part of the story for which it was sold,
[04:51] which is all I wanted, and I think I played that role. So in that sense, I'm happy. As I understand it, you started having a little bit of success with these Euler equations, which are kind of like Navier-Stokes-like,
[05:06] as I would think about it. They're not quite Navier-Stokes, but they're very related. You were having some success with that. When you started having that success and people started seeing the next step maybe being a jump up onto NaviSex.
[05:18] Did you start to think, I could be the guy or I could be one of the guys? Did it suddenly seem possible then? I wouldn't even say it's NaviSex life. It's the main mechanism. So the difference between NaviSex and the oil equations
[05:34] is that the NaviSex equation has this viscosity and viscosity is this internal friction. And throughout my entire life, I've always thought of this viscosity as an annoyance.
[05:48] It's just something that makes the problem a little bit harder. But the mechanism is Euler. So essentially, the difference between solving Euler and solving Navier-Stokes is that you need the singularity for Euler to be a stronger singularity
[06:03] to overcome this internal friction. So absolutely, once we solved Euler, But then we started on this path.
[06:15] Not long after, we saw hyper-discordance. And effectively, you just put the viscosity back in, but you weaken it a little bit. And then it was just like a goal of basically pushing, pushing, pushing, pushing.
[06:29] So we saw, I think I roughly said within a month or so, and it was a clear path. So the other things that we could have done is that you could put an idea of stoics with higher dimensions. So it was basically a clear path forward towards now the idea of stoics.
[06:47] You were a guy who used a lot of AI. You weren't like this Luddite who was anti-AI. You were even using it in this work, weren't you? Absolutely. Yeah, and sometimes the story sort of gets mistold somewhat,
[07:02] but because I'm a mathematician that actually does do pen and paper, Max and comes up with these ideas that, you know, I was just, you know, getting maybe the AI to do some cell check. No, no, no, no.
[07:14] I was working with Laval, who is, like, has himself built this amazing system. Like, people think that if you're just, you push a button, like, you put it into the prompt and you push a button and then out comes the result.
[07:29] And sometimes that's what the AI company is one to sell. is that, like, you don't want to solve it. You just put the question in and you press answer. No, like, I've been working with AI for years,
[07:41] and Lavant himself has built this, like, amazing system, a genetic system, in order to sort of mimic some of the things
[07:54] that us mathematicians do. And it's this system that sort of allowed us to come up with all these, you know, that came up with these great ideas that led to these solutions. As I understand it, people in the field knew that you had this success
[08:09] with the Euler equations, although you hadn't formally published yet, but there was a lot of, there was some buzz starting to go around. Is that right? I wouldn't say that it was necessarily attached to me.
[08:21] So there was a rumour, so there was a, some sort of, there was a leaf from Anthropic and somehow it went to open my eye. And then it went to DeepMind
[08:34] and it went to within the tech industry and then a few days later it came out online. And it was this deep sort of, there was all these tweets. And if you look at the sort of betting market, suddenly it says like,
[08:46] Anthropica's going to sell the Meridian Prize. And I think the rumor was that Anthropica sold two Meridian Prizes. Was NaviAce one of them or was that not, were they not named? Because there are a few. I think the general thing was it was Navier-Stokes and the Hodge conjecture.
[09:00] I think that was the general... I mean, you have to look back, but I think that was the two problems that were... And then it was narrowed to Navier-Stokes. But you don't think that rumor had any kind of genesis
[09:13] in the success you've been having, structurally almost, with the Euler equations? That's exactly where it came from, yes. So that was... Yeah, so there was a leak from Anthropix directly about...
[09:27] It came because people in the project had seen what Levent was doing, and that leaked out. Okay. And then it got misinterpreted, and then more context...
[09:39] That's when the Hodge context got added. But yes, it was directly related to the two of our work. But it was just wrong. The rumor itself was actually wrong, because it said...
[09:51] We hadn't actually thought about it, so it's called oil exploration. And the original rumour was that it was a company, Anthropic. And then they realised it was Levent,
[10:05] and then people from OpenAI started asking Levent, like old friends, started asking, are you in New York? and like eventually they kind of like, you know.
[10:21] And then I got this email from someone from the UK saying, like we heard this rumour that people from Toronto or NYU
[10:33] had inside knowledge that they didn't have these types of problems. Okay. So the rumour had blown up beyond what was the case But it wasn't like completely unfounded because you had taken another step closer to Navier Stokes.
[10:49] And at this point, it sounds like OpenAI jumped in two-footed, as they say, and released the form, released the agents, and then announced what they announced. Before we talk about that, they announced pretty much that Navier Stokes had been cracked,
[11:05] it had been solved, you know, the millennium problem was achieved. Is that the case? Because I'm reading some people saying, well, maybe they haven't, maybe they haven't. And I know it's early days and there's a lot to wade through for people like you, but is
[11:17] it the consensus among people like you that it has been cracked down? Yes, yes. I believe so. Yeah I mean it pretty soft I mean the paper they presented is not like in a readable form but like you know the ideas can be digested
[11:38] and they can be turned into something that's acceptable by a mathematician. So we had this rumor that was being spread around that we had solved another case of money,
[11:50] but at that point it was getting a little crazy, So I reached out to one of the mathematicians from OpenAI to sort of just calm things down.
[12:02] And initially he responded and said, yeah, let's not compete. And I said to them, we're using your product as well.
[12:16] So we wanted... Let's make that clear then, because obviously your collaborator, Levant, is from Anthropic and is using Anthropic products, but you guys are also using OpenAI products. That's right.
[12:28] Well, I'm using OpenAI products, actually. Right. And the point of the messaging that we're using, I wanted to define the situation.
[12:40] I didn't want it to be Anthropic versus OpenAI. I wanted the story to be, this is what you can do at guiding AI. This is what is now achievable.
[12:53] And I think we have to reassess how mathematics goes forward. That was the story. I didn't want it to be, you know, we have such better models
[13:05] and everybody has such better models and so on. I contacted them by email and then we were still writing out papers and so I said that
[13:21] let's meet, I eventually said let's meet the following week and then we got this message from the mathematician that OpenAI was that system in your institution.
[13:37] That was what they said. and we didn't know what it was but we I mean you can make whatever assumptions you'd like to say but like open our eyes
[13:49] about stupid and really stupid and we need to get on a call with us on the Saturday. But stupid sounds like stupid in the context of kind of crazy you know not stupid as in like dumb
[14:02] stupid as in no no yeah stupid as in not stupid as in dumb as if it was stupidism. They're about to do something incredibly unethical.
[14:14] So you did take it, you took it as like, not, hey, we're about to do something wild and crazy. You took it more as we're about to do something we shouldn't do. Absolutely, yeah. So we took it as like, they're about to do something really bad
[14:27] that's going to end, and you need to get on the call to stop this from happening. All right. But at the end of the day, There was no specifics of what they were about to do.
[14:40] Yeah, okay. Okay. And what was the stupid thing? Well, we don't know. I mean, you can't forget anything that was just trying to release before us or something like that. Oh, at the time, they still thought we had solved the Matthew Spokes book.
[14:56] Okay. So, you know, I don't want to guess what the stupid thing they were about to do. but anyway
[15:08] it was enough to convince us to get on the call so so that's when I got on the call is this when they told you they'd solved it or they hadn't done it yet I think Sebastian had like
[15:20] I think maybe I can't remember I mean they had they told us they'd solved it at the beginning of the call but I think like if I recall correctly he had mentioned it
[15:32] maybe in a text to Levent like like a few minutes before the call was over. Okay. And then the first question I had is, like, which wasn't solved?
[15:45] Right. The sort of, the minutiae of, because there were different, yeah, there were different ways of solving it, and there were different, there were a few different finish lines that were set out. So you were just curious technically, from a technical perspective, as a mathematician, you know.
[15:58] Actually, no, it wasn't curiosity. It was, like, I wanted to know if they'd taken our method. Because there was, like, if they had solved the fourth one, then it was suggested, you know,
[16:10] it was like that was kind of a red flag to me that they'd solved the fourth one. And that's the one that they had solved. What made you think that they had used your methods
[16:23] or had access to your methods? Did you straight away think, because you used this codex, which is an AI, which is an open AI product, did you try to think that's how they've got it?
[16:36] They've used that to learn what I'm doing? I wasn't sure what happened. You know, there was lots of possible theories. We don't know what was leaked. We don't know if certain ideas were leaked. You don't need that much.
[16:49] That's what people, you know, like, of course, having the credit sessions is amazing. But you don't need, like, so much if you have enough compute.
[17:02] So it took us a while to figure out what had happened. But initially, the initial distribution is that they started lying. So the first thing that they came in the call was that, and they came prepared for this.
[17:21] This wasn't like they just messed up. They came prepared. They wanted to show us the prompt. and at this point I said I knew that it was about the lie because it lied before about the
[17:33] prompts they used to solve other math problems but then, you know, the math magician showed me on the screen, so this was I think a Google Meet session he showed me on the screen and they just literally copy pasted the Millennium Prize problem
[17:47] so they said that they took the prize and they just entered and they got a solution and at that point I just had to laugh because it's so ridiculous It came out within a few minutes that this was a lie and that they had worked on other
[18:01] problems beforehand and they had given other prompts and they... This lie evaporated within minutes, but it's what they came prepared to give me. And also they wouldn't tell me when they started working on it.
[18:14] I kept asking them, when are you going to start working on it? And they just said, oh, we have these amazing models, we have these amazing models. I'm not someone who, like, I'm someone in the field that understands how this works
[18:30] and understands that model is one thing, like the number of agents and stuff like that. There's more to the story there. And eventually I said, I sort of did a loaded question to Sebastian.
[18:44] I said, so you agree that you only started working on it after RIMA? And he said, yes. I mean, there's no law against snap interaction based on a room you've heard. If I heard a rumor someone was going to make a YouTube video that I was about to make,
[18:58] I'd probably put mine on the express path as well so that I got in first. But do you have any knowledge or belief as to whether or not the stuff you fed into OpenAI,
[19:11] into Codex, was used by OpenAI, whether directly, just directly looking at it, or used to train the model, which I think is not that big a difference anyway,
[19:23] but what do I know? What do you think or what do you know? So I can't do 100% pinpoint exactly what was happening, but the circumstantial evidence is clear.
[19:37] So everyone keeps pointing to the NaviFresh problem and comparing it to our work, But that's not the right paper to compare to. The right paper to compare is the unforced Euler equation.
[19:52] Now, the unforced Euler equation, all the mechanisms in this unforced Euler equation are identical to a different version of the Euler paper that we haven't released yet.
[20:04] Like, they themselves are based on all the ideas of the paper that we did release. And the steps to get from forced to unforced to Euler was actually very small.
[20:19] Basically, all the main architectural ingredients in that unforced to Euler were in our work, for which they would have had on their service.
[20:32] And it hadn't been published. It hadn't been published, no. No. It hadn't been published. And the other point that's in the context is that OpenAI have admitted that, not for our project,
[20:44] that their agents have actually even gone so far as hacking another competitor's GitHub repository in order to solve a math problem.
[20:58] So I think it was just a few days ago that they announced that agents were working on some math problem, and then they realized that there was another team working on it
[21:10] and then they managed to get the token which allowed them access to the GitHub repository of their competitors in order to solve the problem. I mean, there's so many ways in which they could get access to work.
[21:28] And the key thing here is that to get to the map is such a problem they used 10,000 agents. They didn't use 10,000 agents to get to the order problem. They used 100 agents.
[21:40] So they worked from absolutely nothing with only 100 agents. And that is key, the number of agents, because it's called an E. A number of agents means like how wide you can search.
[21:53] And they used 100 agents, and then they end up with an identical architecture to what we had. Do you regret using their product? Were you naive? Because you know this industry.
[22:05] You said yourself you know the industry. You know how it all works. Did you make a mistake by putting all that stuff onto Codex? Do you regret it? Would you do it again? Yeah, we made many mistakes. I mean, you know, there was a mistake in what caused the leak in the first place.
[22:23] I mean, you know, there was many, you know, had there never been a leak, then we wouldn't have had this red alert. But have I never used Kodak? Absolutely. Yes, I guess we were.
[22:36] You know, I didn't think that... I've seen a lot of crazy stuff. I didn't think that they would go this far. Do you think you should have been more secretive? I've heard stories about Andrew Wilde from Ars La Sim,
[22:48] and he was very secretive about what he was working on. No one knew he was doing it until he stood in front of that blackboard. Were you not secretive enough? Is that the problem? We thought we were But I mean the I mean the leak was the leak was bad right That was
[23:07] that was, that was the big mistake. Was that because of computers, or was that just humans talking over coffee at the water cooler? That's
[23:19] exactly, I think that's what happened. I mean, I don't actually know exactly. There's a lot of, like, conflicting stories or whatever, but I feel like it was people bragging about what their company is doing
[23:35] and stuff like that. I've read your statement, and I'll link to it again so people can really look at it in depth, but you make it pretty clear in that statement that you think you probably would have got more credit attribution and been more involved if your collaborator didn't
[23:51] happen to be an employee of Anthropic. Is that the case? Do you think half the reason that you've been put in the position you're in is because of this rivalry between OpenAI and Anthropic?
[24:04] Well, I want to push back on the credit thing because I'm not after having more credit for the work. It's just I don't like what I'm trying to push back
[24:16] in the culture of how these companies are acting and the negative effect it is having on the mathematical community.
[24:31] So it's less about trying to add more credit or having... I don't care about receiving the Clayton-Milling Prize. The problem is that this ultra-competitiveness
[24:46] between the labs. And I should say, they were the only lab that started spending millions of dollars to solve the Navier-Fox problem after the RUMAC.
[24:58] And so I said, that's another well-known company that did the same. So, like, the... What I want to push back on is, like,
[25:11] the way that this new past, great culture is like head-on collision with the mass community and in the future will be a head-on collision with the rest of the
[25:24] past as well. Other than the way that you and your collaborator and a few other people have been treated and kind of the lack of manners and the lack of propriety,
[25:36] other than that, isn't this a good thing? Isn't this crazy competition and all the resources they've thrown into this, advancing the field, I mean, they cracked this problem that you all dreamed of cracking,
[25:49] which they may not have done. I mean, it could have been decades before this problem was solved and it got done really quickly. As a mathematician who I presume is like, you know,
[26:01] in this pursuit of truth and more knowledge, like this has actually moved the frontier. So let's look at what happens if they hadn't done it. if they would have cracked it a few days later.
[26:15] The world would not have changed. And now let's look at what the negative impacts have had on it. Like, people are, you know, like, had the Millennium Prize, you know,
[26:27] problem been solved on September 8th, I think it was, and not, like, September 20th, the world would not have changed. Now, it has had huge negative impacts on the mathematical community because now everyone
[26:43] doesn't want to talk about their open problems. People are afraid that these AI companies will suddenly come in and treat them as benchmark
[26:55] problems. don't trust any of these AI products. It's just like it's had a huge negative impact on academia.
[27:11] It sort of slowed us down because of the threat that these AI labs pose on us. So it's actually had a negative impact on scientific progress.
[27:25] And also the matter of releasing a result a few weeks earlier. Now, I love the Millennium Prize, but this is not curing cancer.
[27:40] And having solved this Millennium Prize, like September 8 or September 20, would have zero impact on the world. You said yourself, though, that you had heard that another AI company was working on it and had thrown all the resources at it.
[27:54] which, you know, you can think what you want about that, but it does show why OpenAI would have maybe been keen to get it out there those few days earlier. It makes all the difference to them because they get to say we did it or they did it, like, you know.
[28:07] And they're competitive. Mathematicians are competitive too. Like, everyone here has a competitive streak. Yeah, for them it's a matter of, I mean, I don't think it's as important as they think it is, but for them it's like a matter of, like, life or death of their product.
[28:22] They think that had Antropic released a millennium-sized problem, they think that would have affected their IPO. Like, it shows that Antropic had much better internal models of Navy.
[28:45] But that is the rationale behind it. So that was the big threat to them. it was a threat to their bottom line, that it would show... And it's kind of missing the point altogether
[28:57] because it isn't just about the internal model. It's about using a lot of them. And it's like you need to use them... You know, there's one part of the story...
[29:12] And I think I saw an open... I am quite making this claim that it's all about internal models. like 90% of internal models and 10%...
[29:25] It's the opposite way around. It's maybe 20, 30% of the internal model, and then the rest of it is using all these different agents
[29:37] within a particular harness or a particular framework in order to solve a problem. So you've made something of a stand here.
[29:49] You've been willing to speak to people like me. You've made public statements. This is a huge, big, rich company with a lot of power and resources. Was that a difficult decision for you, like, to be principled about this?
[30:01] You could easily have gone another direction here. Like, did you think about consequences? Have you thought much about that, or has this just seemed like the natural thing to do? I mean, you know
[30:13] I was offered the opportunity to write as sole author of their paper and throw my collaborator under the bus now, like, did I
[30:27] think about the opportunity of being sole author? Not for one second, I mean that's it's so unethical. They were asking me to throw him under the bus
[30:40] someone who I'd been working with for a year for throwing it to Angela just because he's an anthropic employee, that was the only reason the reason they said it's so annoying that
[30:52] Levant is an anthropic employee otherwise they would have liked Adam as the author, so the sole reason they didn't want to find a solution with Levant
[31:06] was because he was an anthropic employee these 10,000 agents this model that cracked Navier-Stokes was it a moment of genius? Do you look at how it was done and think
[31:19] you know that was pretty amazing that was pretty smart like well done you Yeah so I mean I've been giving these media interviews quite a bit
[31:31] so I haven't had so much time to digest the proof. But from what I've seen, I'll talk about, I think I'll talk about this
[31:43] at a later date. But the key idea that went from the Euler to Navier-Stokes is a smart idea. And from what I can tell,
[31:55] it's not what they are talking about in the introduction of the paper. And it's not what's actually being broadly talked about I find that actually quite amazing
[32:07] that people are missing what the key idea was to get from oil and earth and that kind of stuff Can you even give us a hint? I know you haven't fully digested it yourself yet but as someone who made a video about it I'd love to
[32:19] know what we've all missed Yeah so what everyone has been explaining is that the vortex, this collapsing
[32:31] vortex And this is the principal idea of what causes the solutions for Mavie and Pryor. This collapsing vortex is not a solution.
[32:45] It's not a solution to Mavie and Pryor. So you can create endless non-solutions to the Mavie and Pryor surprise when you have
[32:57] four things, but you just put everything in the vortex. Like, you make a mistake, and whatever the error is, you say the force equals the error. So the collapsing vortex is not a solution.
[33:13] So if you were to present that, it would have an infinite force. So you would just break putting infinite energy into a solution, and that's not surprising. Anyone can do that.
[33:26] Okay. The secret is how do you take something which is not a solution and turn it into a solution without having an infinite fortune?
[33:39] So how do you correct that? How do you fix this non-solution? And this is where the key idea comes into. And the key idea is to use a concept called convex integration, which I'm surprised it's
[33:58] what I've built my whole career on. But to use combined ideas from convex integration with this growth mechanism from the Euler
[34:13] So it's to combine some techniques from complex integration. And complex integration is kind of like a way of fixing things.
[34:26] It a mechanism of fixing the error It used in a different way in this group but it traces its history all the way back to John Patch Now you want to combine these ideas of convex integration
[34:43] with the growth mechanism of the Euler's law. And if you combine these two ideas, you create
[34:55] this new mechanism which is used to correct this non-solution. This is actually a cool idea. And it's actually something that... Some sort of idea that...
[35:12] It's kind of an idea that I've been trying and failing to do for, like, I don't know, for over ten years or more. to be able to use these complex integrations.
[35:26] And in fact, even with my post-credit for the last year, I've been trying to combine these ideas of complex integration with these ideas of solution. There you go. I'm not saying that I didn't achieve it.
[35:38] I didn't manage to do it. Have you looked at the open IO paper yet in enough detail to look at that and think, ah, yes? Maybe you would have done it in 10 years. Maybe you wouldn't have. But have you been able to look at it yet and think,
[35:50] I see it now, I see the leap. Is there that moment? This is the leap. So this was the leap. I haven't had enough time to fully... I literally have only spent a couple of days looking at it.
[36:06] But this was the leap. One thing that's kind of... When AI companies just release the stock and they didn't fight any of the... They did a terrible job of fighting.
[36:19] They did actually cite my paper with Dr. Cole on context integration, and that was like an early hint, but they didn't actually say in which, how it was used, that paper.
[36:32] They did cite a different paper by Sarah Danieri, and that was technically, technically with my supervisor, where they did provide like a miniscule idea that they used this mechanism.
[36:48] This controversy that, unfortunately, you found yourself very much at the centre of has become this kind of kernel of this kind of push against AI.
[37:02] We've seen this petition signed by all these field medalists, Terry Tao and the likes, that are firing off these warning shots about AI being brought in onto these big problems. But you're an AI guy.
[37:14] It feels like to me you're an AI guy. You're like, you think you've been on board with this. How do you feel about that petition? And how do you feel about your role in its birth and things like that? Are you on board with the petition?
[37:26] Do you think they've overreacted? Tell me how you feel about that. I think, you know, I don't want to sway, you know, I don't want to describe what's correct and what's wrong. I think that we are in a new world in terms of what AI can do,
[37:46] and I think we should hear everyone's voices, and I think that we need to hear conflicting voices
[37:59] in order to figure out a good path forward. Now, as I said, I am someone who uses AI extensively, and I have been using it for years now.
[38:11] You know, a funny story myself is that I came from... I was a computer scientist originally. So when I was 13 years old, I was writing... I used these weird things called GPUs, which
[38:28] It has to be something important to write computer graphics. Computer graphics engine when I was 13. And so I wanted to be a computer game programmer.
[38:40] And so I had this temptation, at some point, you know, I had an uncle who was a mathematician, I think he was, and he was looking to go into math, and I went into pen and paper, and I was literally just writing everything in pen and paper.
[38:53] And then I sort of came back and I became, you know, using AI to do the system. So I, look, I don't think you can go back. I don't think we can, you know, mathematics is not a form of this science.
[39:09] We, you know, I mean, there's an analogy between Deep Blue and Cash Club, but the big difference there is that we're quite happy to watch human players play chess.
[39:23] So far, I haven't seen an interest of putting two mathematicians in a room and having them lie solve a math problem. That's not...
[39:35] I don't think that's great entertainment. So I think we have to... I don't think we can go back. I think we have to find a positive way forward that does incorporate AI.
[39:50] But, yeah, obviously, this incident with Argonautics is that there's a lot of things that need to be fixed. Because you're such an AI guy and have been so good at using it, it does have this whole flying too close to the sun kind of feel to it.
[40:06] Do you see that? Yeah, so, you know, one thing I often say is that, like, this was not the result we had. It was not our only result.
[40:18] I have this WhatsApp group with Levent, and we have a whole bunch of results with different problems. One thing I could do is pretend nothing has changed and just release all these results
[40:32] and get all my publications in the top journals and pretend. But I think that's silly. I think some people might think, oh, you know, releasing, holding back the truth is not a good thing.
[40:48] But, I mean, we have to be real here, so it's not going to change the world if we don't release the pure math truth. So I think we need to, I think there's more important things here, and we have to take a step back.
[41:01] and besides I think people would be more interested in knowing how we solve the problem than
[41:13] than just using our technique secretly to solve more problems and the responsibility to explain our methods and explain how we went about
[41:25] solving these problems and that's what I think I'll dedicate my time to in the next few months. If you had found out that a mathematician or a couple of mathematicians in Japan were pretty
[41:37] much on the same track as you, you found out via a rumor or a call, and then you and Levant had said, okay, we're going to put our foot from the accelerator to make sure we publish before these guys, and then you did.
[41:50] How is that different to what OpenAI did, other than the fact they can do it so much more quickly, like inhumanly quickly? I want to really nail down what OpenAI, because you said if open-hour, did this a few days later, you'd be cool with it. What's the thing they did
[42:05] that you think's really bad? So in that circumstance, say someone in Japan had this problem and then I was working with an event and he had access to internal models, I would not try to outrun them. I think that's
[42:26] right there. Right. Right, I think if there was a group of mathematicians, like let's say that Lewis and Diego, whose work I built up, were close to solving unforced Euler or close to solving Magnus Stokes.
[42:44] had I had that impression then there's no way I would have had used resources that they
[42:56] don't have access to to front run them. I mean I think that's inappropriate. That's the main thing, the kind of the code of conduct of mathematicians is if someone else is on a
[43:08] similar place to you you sort of join forces rather than accelerate to defeat them I don't want to say there's a great code of conduct, because there's a lot of famous
[43:21] incidents in mathematics itself where people have presented some work or talked about some work or presented it at some conference and then they went off and wrote down notes and
[43:35] then tried to front them. Now this is frowned upon, absolutely frowned upon within the mathematical community, but it does happen. speaking to you as a fellow Australian, some of the attitude it feels like you're taking
[43:47] is like cricket-like. It's just not cricket to do that. It's just not the right, it's not the spirit of the game to do it. And that's fair enough, and I respect that more than anyone, but you can also be ruthless if you want to be. But I would say that
[44:00] I think if it had just been that, you know, I would have dealt with it. And, like, I think it's still wrong.
[44:12] that it wouldn't be. But that's not what I believe happened. If they've gone into... I mean, they'll say they haven't, you know. But if they've gone into Codex and taken your work out of that
[44:26] that you put in there in good faith, that does feel like you've had your pocket picked. Yeah. And that's why I'm upset. All right. And I just want to be clear
[44:38] because I don't know if they're going to speak to me. They say that they will claim I mean, they have claimed to cover him multiple times, and I will link to it, but that's not what they did. Absolutely. But they also said that they one-shotted Navier's books.
[44:52] Will this change the way you use AI as a mathematician? It certainly changed the way I use codecs. Nothing important goes into codecs.
[45:04] So, nothing of any actual value. Any actual original ideas will go into CodexMath.
[45:34] that makes complete sense, we've got that bit right, there's no reason for that not to be correct but then the mathematical complexity of it, the tricky bit is
[45:46] we don't know if the truth is always going to exist and so we kind of find ways to cheat so we might make simplifications, we might make assumptions to reduce some of these terms or to remove time from the problem
[45:59] you can get ways around it by making assumptions and making simplifications it.